22 · Audio

Tuesday, Nov 10, 2026

Materials for this session are not published yet. They appear here before class.

Objectives

By the end of this session you can:

• Transcribe speech with timestamps and speaker labels.
• Extract the metadata that means something: pauses, disfluencies, tone.
• Build the utterance table.
• Audit a transcript by sampled listening, with a word-error rate.

What we cover

• What audio has measured: the Fed chair’s vocal tone moves markets; the affect in earnings calls predicts.
• Whisper-class transcription with timestamps and speakers.
• Past the words: pauses, disfluencies, pitch and energy, emotion labels — audio is text with meaningful metadata.
• The deliverable shape: one row per utterance — speaker, text, start and end, pause before, tone.

Verification habit. Spot-listen to a sample; report word-error rate on a hand-checked subset.